Velez-Irizarry et al. BMC Genomics (2019) 20:3 https://doi.org/10.1186/s12864-018-5386-2

RESEARCHARTICLE Open Access Genetic control of longissimus dorsi muscle gene expression variation and joint analysis with phenotypic quantitative trait loci in pigs Deborah Velez-Irizarry1, Sebastian Casiro1, Kaitlyn R. Daza1, Ronald O. Bates1, Nancy E. Raney1, Juan P. Steibel1,2 and Catherine W. Ernst1*

Abstract Background: Economically important growth and meat quality traits in pigs are controlled by cascading molecular events occurring during development and continuing throughout the conversion of muscle to meat. However, little is known about the genes and molecular mechanisms involved in this process. Evaluating transcriptomic profiles of skeletal muscle during the initial steps leading to the conversion of muscle to meat can identify key regulators of polygenic phenotypes. In addition, mapping transcript abundance through genome-wide association analysis using high-density marker genotypes allows identification of genomic regions that control gene expression, referred to as expression quantitative trait loci (eQTL). In this study, we perform eQTL analyses to identify potential candidate genes and molecular markers regulating growth and meat quality traits in pigs. Results: Messenger RNA transcripts obtained with RNA-seq of longissimus dorsi muscle from 168 F2 animals from a Duroc x Pietrain pig resource population were used to estimate gene expression variation subject to genetic control by mapping eQTL. A total of 339 eQTL were mapped (FDR ≤ 0.01) with 191 exhibiting local-acting regulation. Joint analysis of eQTL with phenotypic QTL (pQTL) segregating in our population revealed 16 genes significantly associated with 21 pQTL for meat quality, carcass composition and growth traits. Ten of these pQTL were for meat quality phenotypes that co-localized with one eQTL on SSC2 (8.8-Mb region) and 11 eQTL on SSC15 (121-Mb region). Biological processes identified for co-localized eQTL genes include calcium signaling (FERM, MRLN, PKP2 and CHRNA9), energy metabolism (SUCLG2 and PFKFB3) and redox hemostasis (NQO1 and CEP128), and results support an important role for activation of the PI3K-Akt-mTOR signaling pathway during the initial conversion of muscle to meat. Conclusion: Co-localization of eQTL with pQTL identified molecular markers significantly associated with both economically important phenotypes and gene transcript abundance. This study reveals candidate genes contributing to variation in pig production traits, and provides new knowledge regarding the genetic architecture of meat quality phenotypes. Keywords: eQTL, Skeletal muscle, RNA-Seq, Transcriptome, Pig

* Correspondence: [email protected] 1Department of Animal Science, Michigan State University, East Lansing, MI 48824, USA Full list of author information is available at the end of the article

© The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 2 of 19

Background where a single marker is associated with the expression of Genomic improvement techniques have significantly multiple genes, serving as a potential master regulator that advanced livestock breeding in recent years. Genomic can account for a significant portion of phenotypic regions harboring single nucleotide polymorphisms variation. A co-localization analysis of eQTL with pQTL (SNP) accounting for a significant portion of phenotypic reveals novel insights into the genetic architecture of meat variation for economically important traits have been quality, carcass composition and growth traits. identified and implemented in marker assisted selection [1–3]. In pigs, these efforts have identified candidate Results genes affecting meat quality (e.g., CRC1, PRKAG3, Identification of eQTL CAST), weight gain (e.g., MC4R) and litter size (e.g., A genome wide association study (GWAS) was con- ESR)[4]. However, we still do not fully understand the ducted using 23,162 SNP markers and 15,249 transcript molecular mechanisms underlying the variability observed abundance profiles for 168 F2 pigs. The GWAS identi- in pork traits. fied 339 eQTL (3094 significant gene marker associa- Meat quality traits are highly correlated. During the tions; whole genome FDR ≤ 0.01 per gene) for 321 gene conversion of muscle to meat, Ca2+ ions are released from transcripts and 2523 molecular markers (Additional file 1 the sarcoplasmic reticulum and the anaerobic production Table S1). The number of SNP associated with an eQTL of ATP leads to the accumulation of lactic acid that reduces was on average 9.09 ± 15.07, and the size of each eQTL muscle pH [5].TherateofpHdeclineandreleaseofCa2+ interval was on average 11.59 ± 22.30-Mb (Table 1). A directly influences water holding capacity, meat color and total of five eQTL intervals contained an additional peak the rate of proteolytic activity that leads to meat tenderiza- determined through conditional analysis by fixing the tion [5]. While these molecular processes have been peak eQTL SNP. These eQTL intervals were observed extensively studied with numerous QTL identified for ten- on SSC1, 10 and 12 (entries in Table S1 with the same derness, drip loss, pH, meat color and enzyme activity [6], transcript identifiers). we know little of the genetic architecture regulating these All autosomes had associated eQTL, with SSC9 contain- traits. This is likely due to the high variability of meat qual- ing the most associations (42 eQTL). We considered a single ity traits that are known to be heavily influenced by both marker associated with more than ten genes to be a plaus- genetic and environmental factors such as antemortem ible putative hotspot, and two chromosomes contained a handling [7–9]. Regulators of gene expression have been putative hotspot; SSC9 (ASGA0044684; SSC9:125.0-Mb) used to study the molecular bases of polygenetic pheno- and SSC15 (H3GA0052416; SSC15:121.8-Mb). ASGA0044 typic differences in swine populations [10–13]. In addition, 684 was associated with 25 transcripts, and H3GA0052416 expression quantitative trait loci (eQTL) maps provide a with 11 transcripts (FDR ≤ 0.01). Both plausible hotspots foundation to study divergent molecular processes in live- mapped to non-coding regions, an intron variant of the ral stock species [2, 14]. This approach has been successful in guanine nucleotide dissociation stimulator like 1 (RGL1) identifying candidate genes, causative variants and molecu- gene on SSC9, and an intergenic variant on SSC15. lar networks regulating phenotypic traits in swine, includ- ing back fat [15], drip loss [16], glycolytic potential [13], Local versus distant regulators of gene expression plasma cortisol levels [10] and lipid metabolism [17]. For each of the eQTL intervals, a plausible position For meat quality traits, cascading molecular events range delimited by the first and last significant marker starting before exsanguination and continuing through- (FDR ≤ 0.01) was identified and compared to the mapped out the conversion of muscle to meat play a critical role position of the associated gene transcript to distinguish in determining the eating quality of pork. By studying between local and distant regulators of gene expression the transcriptomic profile of the initial steps leading to (Fig. 1 and Additional file 2 Figure S1). A classification the conversion of muscle to meat, we can elucidate key of local-acting regulator of gene expression was deter- regulators of polygenetic trait phenotypes. Specifically, mined if the position of the associated gene transcript we can identify gene transcripts subject to genetic con- overlapped the eQTL interval (Additional file 2 Figure S1). trol that potentially regulate complex traits by mapping We identified 166 local regulators of gene expression eQTL and testing their co-localization with phenotypic (Fig. 1,blackassociations). QTL (pQTL). The average distance from the gene position and peak In this study, we use an F2 Duroc x Pietrain resource eQTL SNP for local regulators was 1.90 ± 3.86-Mb. population developed at Michigan State University [18, 19] However, due to the large plausible position range for (the MSUPRP) to map eQTL for longissimus dorsi muscle some local eQTL (up to 175-Mb), the maximum dis- in order to identify local and distant regulators of transcript tance for a local regulator was 25-Mb (Table 1). If the abundance, and to estimate narrow-sense heritability (h2) gene mapped to the same chromosome but fell outside of gene expression. Putative hotspots are also of interest the range of its associated eQTL with markers below the Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 3 of 19

Table 1 eQTL summary among regulator types Gene Regulator Na Minb Maxc Meand SDe Average length of eQTL interval f All regulators 339 0 175.19 11.59 22.30 Local 168 0 175.19 21.62 27.59 Plausible Local 23 0 11.44 1.36 2.77 Distant Same Chromosome 61 0 25.55 2.22 5.46 Distant 87 0 69.76 1.47 7.92 Average distance from eQTL to gene transcript position f Total Same Chromosome 252 4.11e-4 104.75 3.65 12.15 Local 168 4.11e-4 25.07 1.90 3.86 Plausible Local 23 4.28e-3 1.49 0.20 0.40 Distant Same Chromosome 61 3.73e-3 104.75 9.75 22.92 Distant 87 –––– Number of SNP associations per eQTL All regulators 339 1 105 9.09 15.07 Local 168 1 105 16.45 18.65 Fig. 1 eQTL map. The y-axis represents the absolute genomic Plausible Local 23 1 14 2.87 3.08 position of the gene and the x-axis represents the genomic location of its associated SNP marker. Associations aligning on the diagonal Distant Same Chromosome 61 1 17 2.05 2.37 are eQTL found on the same chromosome as the gene. A plausible Distant 87 1 5 1.46 0.97 position range was identified for each eQTL interval based on the peak’s flanking markers, and local regulation was determined when Heritability estimates the gene position overlapped this range, shown in black. Plausible All regulators 339 5.47e-10 0.97 0.32 0.23 local regulators of gene expression (described in Figure S1 in Local 168 5.47e-10 0.97 0.42 0.22 Additional file 2) are shown in yellow. The eQTL intervals shown in green are distant regulators that map to the same chromosome as Plausible Local 23 0.04 0.63 0.32 0.16 their associated gene. Distant regulators mapping to a different Distant Same Chromosome 61 1.19e-09 0.78 0.29 0.23 chromosome than the associated gene are shown in blue. The eQTL shown in red are plausible putative hotspots on SSC9 and SSC15 Distant 87 1.34e-09 0.76 0.17 0.17 aNumber of eQTL bMinimum value regulator (Additional file 2 Figure S1). We observed 87 cMaximum value dAverage value distant regulators of gene expression (Fig. 1, blue associ- eStandard deviation of value ations). A non- parametric test showed local eQTL had fValues of eQTL interval or distance shown in mega bases. Zero interval values correspond to eQTL associated with a single SNP significantly higher numbers of associated SNPs than distant eQTL (p-value ≤2.20e-16). significance threshold between the gene and eQTL posi- tions, the eQTL was considered to be a distant regulator Heritability of gene expression on the same chromosome as the associated gene Heritability (h2) was estimated for all gene transcripts (Additional file 2 Figure S1). A total of 61 distant regula- with 344 exhibiting significantly heritable expression tors on the same chromosome as the associated gene (FDR ≤ 0.01, p-value ≤2.27e-04). The mean h2 for tran- were identified (Fig. 1, green associations) with their scripts with significantly heritable expression was 0.51 ± eQTL interval at an average distance of 9.75 ± 22.92-Mb 0.13, whereas the mean h2 for other transcripts was 0.09 from the associated gene position (Table 1). However, in ± 0.12 (Table 2). The relationship between the estimated situations where the area between the eQTL range and h2 of gene expression and its significance is shown in the associated gene transcript were found to be devoid Figure S2 in Additional file 2. Significant enrichment of of markers, the eQTL was considered to be a plausible genes associated with an eQTL was observed for the sig- local regulator (Additional file 2 Figure S1). Under this nificantly heritable gene transcripts (p-value ≤2.2e-16; classification, 23 plausible local regulators of gene ex- shown in red, Figure S2). The h2 of genes with an associ- pression were identified (Fig. 1, yellow associations) with ated eQTL that were not significantly heritable was on their eQTL interval at an average distance of 0.20 ± average 0.21 ± 0.16 (shown in yellow, Figure S2 and sum- 0.40-Mb from the associated gene position (Table 1). An marized in Table 2), whereas the group of significantly eQTL that mapped to a different chromosome than its heritable genes associated with an eQTL had a mean h2 associated gene transcript was classified as a distant of 0.57 ± 0.15 (Table 2). Mean heritability among the Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 4 of 19

Table 2 Heritability summary for all genes and genes with an QTL for 10th rib backfat from 13 to 22 weeks of age, associated eQTL seven QTL for last rib backfat from 13 to 22 weeks of Significant N Heritability (h2) age, one QTL for loin muscle area at 16 weeks of age, 13 h2 Min Max Mean SD QTL for carcass composition traits and 16 QTL for meat All Genes quality traits. Yesa 344 0.184 0.968 0.508 0.133 Meat quality traits in our population exhibited pheno- typic correlations as expected. WBS was negatively corre- No 14,879 2.210e-19 0.785 0.091 0.123 lated with sensory panel scores (i.e., juiciness, tenderness eQTL Genes and overall-tenderness) and cook yield, and positively cor- Yesa 103 0.184 0.968 0.574 0.147 related with protein percent (p-value ≤8e-05, Additional No 218 5.475e-10 0.745 0.206 0.165 file 2 Figure S3). Cook yield was negatively correlated with aFDR ≤ 0.01 drip loss, and positively correlated with 24-h pH and pro- tein percent (p-value ≤8e-05, Figure S3). Phenotypes re- lated to tenderness were associated with QTL on SSC2, different regulator types was higher in the group of and all eight of the aforementioned correlated meat qual- eQTL associated with local-acting regulation, 0.42 ± ity phenotypes were associated with QTL mapped to 0.22, and lowest in eQTL with distant-acting regulation, SSC15 (Fig. 2). A similar trend was observed for growth 0.17 ± 0.17 (Table 1). A non-parametric test showed a and carcass composition traits related to fat deposition significant difference between heritabilities for local- and and muscle weight where serial ultrasound measures for distant-acting regulators (p-value ≤1.08e-14). 10th and last rib backfat were positively correlated with carcass 10th-rib and last lumbar backfat, and negatively Phenotypic QTL correlated with loin weight (p-value ≤8e-05, Figure S3), Genomic regions significantly associated with growth and these traits were associated with QTL on SSC6 (Fig. 2 [20], meat quality and carcass composition [21] traits and Fig. 3). have been previously identified in our MSUPRP. How- ever, these analyses used an earlier assembly of the pig Co-localization of phenotypic QTL with expression QTL genome (Sscrofa10.2); therefore, we reanalyzed the 67 The association of eQTL co-localized with pQTL was phenotypic traits for the F2 population (960 animals) performed through a conditional analysis of transcript following previous methods21,22 to generate an updated abundance, which fixed the peak pQTL SNP, to elucidate QTL map using the most current genome assembly eQTL significantly associated with phenotypic traits. (Sscrofa 11.1). Our QTL analysis of 29 growth traits Manhattan plots of eQTL co-localized with pQTL are identified 14 pQTL (Table S2 in Additional file 1, FDR ≤ shown in Fig. 2 for meat quality and carcass composition 0.05, p-value ≤2.50e-04) for which seven were confirmed traits, and Fig. 3 for growth traits. The conditional ana- from Duarte et al. [20] and five exhibited a different lysis tested 53 eQTL (orange associations) co-localized peak SNP, in part because one of the SNP on SSC6 with 34 pQTL (blue associations) for ten growth and 11 (ALGA0122657) did not have a genomic position in the meat quality and carcass composition traits (Fig. 2 and new genome build. We were unable to confirm two Fig. 3; Table 3 and Additional file 1 Table S3). A total pQTL on SSC2 for 10th rib backfat at 16-weeks and last of 16 eQTL were significantly associated with 21 rib backfat at 19-weeks, and one pQTL on SSC3 for pQTL, where conditioning upon the peak pQTL birth weight that were reported in Gualdrón Duarte et marker resulted in the complete removal of eQTL al. [20]. However, we identified two new pQTL for loin significance (p-value ≤5.95e-04 for SNP effect and muscle area at 16-weeks on SSC6 and last rib backfat at FDR ≤ 0.01 for eQTL significance; black associations 10-weeks on SSC12. Our QTL analysis for carcass com- in Fig. 2 and Fig. 3;Table3 and Table S3 in Additional file position and meat quality traits identified 29 pQTL 1). Three pQTL regions common among correlated pheno- (Additional file 1 Table S2, FDR ≤ 0.05). Fourteen pQTL types co-localized with eQTL, resulting in eQTL sig- were confirmed from Casiro et al. [21] and eight exhib- nificantly associated with variation for multiple ited a different peak SNP, in part because three SNP phenotypes. Pearson correlation between phenotypic (SSC6: ALGA0122657, SSC11: M1GA0015491 and values and the colocalized gene expression resulted in SSC15: MARC0047188) did not have genomic positions 11 genes with expression significantly correlated with in the new genome build. Seven new pQTL were identi- phenotype; 73% of these genes were correlated with fied for cook yield (SSC5 and SSC8), last lumbar backfat protein percent (Additional file 1 Table S3). The LD (SSC4, SSC9 and SSC10), dressing percent (SSC11) and between peak QTL SNP for pQTL significantly loin weight (SSC11). In total, 43 pQTL were mapped colocalized with eQTL was on average 0.317 ± 0.08 using the Sscrofa11.1 genome assembly, including six (Additional file 1 Table S3). Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 5 of 19

Fig. 2 Manhattan plots of meat quality and carcass composition pQTL co-localized with eQTL. The x-axis is the absolute genome position in mega-bases. The y-axis is the negative base 10 logarithm of q-values, with the red line representing the significance threshold. Manhattan plots in shades of blue are for the pQTL (FDR ≤ 0.05), and those in shades of orange are for the eQTL (FDR ≤ 0.01). SNPs associated with an eQTL co-localizing with a pQTL, and whose association is no longer significant after performing the conditional analysis are shown in black

Phenotypic QTL for growth and carcass composition backfat at 13 weeks of age, and carcass last lumbar backfat. traits associated with eQTL on SSC6 revealed two genomic The peak pQTL marker for loin muscle area at 16 weeks of regions. A 28.82-Mb region (SSC6:43.819–72.625-Mb) was age, ASGA0105067, accounted for 4% of the phenotypic associated with the hepsin gene (HSN) and with loin variance and 13.5% of the gene expression variance with in- muscle area at 16 weeks. A 53.33 Mb region (SSC6:99.932– creased loin muscle area associated with decreased expres- 153.261-Mb) was associated with a novel transcript sion of the HPN gene (Fig. 4). The pQTL marker for (SSC6:104.08) and with serial ultrasound measures of last backfat deposition, ALGA0104402, accounted for 5–7.1% rib backfat (at 10, 13, 16 and 22 weeks of age), 10th rib of the phenotypic variance, and 10.1% of the gene Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 6 of 19

Fig. 3 Manhattan plots of growth pQTL co-localized with eQTL. The x-axis is the absolute genome position in mega-bases. The y-axis is the negative base 10 logarithm of q-values, with the red line representing the significance threshold. Manhattan plots in shades of blue are for the pQTL (FDR ≤ 0.05), and those in shades of orange are for the eQTL (FDR ≤ 0.01). SNPs associated with an eQTL co-localizing with a pQTL, and whose association is no longer significant after performing the conditional analysis are shown in black expression variance, with increased expression of the novel rib backfat (M1GA0008917) accounted for 12.2% of the transcript SSC6:104.08 associated with reduced backfat de- phenotypic variance, with increased expression of both position (Fig. 4). the SSC6:104.08 novel transcript and SSX2IP associated Two additional pQTL for carcass composition pheno- with reduced 10th rib backfat. For loin weight, the peak types (carcass 10th rib backfat and loin weight) also pQTL marker (ASGA0029651) was associated with re- mapped to the 53.33-Mb region on SSC6 and were signifi- duced loin weight and reduced expression of both the cantly associated with the SSC6:104.08 novel transcript SSC6:104.08 novel transcript and SSX2IP, accounting for and with SSX2IP. The peak pQTL marker for carcass 10th 6.4% of the phenotypic variance and up to 12.7% of the Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 7 of 19

Table 3 Phenotypic QTL co-localized with expression QTL Phenotype SSC Peak SNP Position Ea VSb h2 qvaluec Intervald Ne Carcass 10th-Rib Backfat 1 ALGA0010839 270.61 + 0.03 0.45 4.05e-03 270.39–274.20 1 WBS 2 M1GA0002229 4.34 – 0.04 0.26 4.07e-04 4.34–7.32 1f SP Tenderness 2 H3GA0005676 6.77 + 0.05 0.29 2.02e-04 4.34–7.32 1f SP Overall Tenderness 2 H3GA0005676 6.77 + 0.05 0.28 4.80e-05 6.38–7.32 1f Last Lumbar Backfat 4 ASGA0092651 88.28 – 0.04 0.41 4.53e-02 0 4 Last-Rib Backfat 16-wk 5 ALGA0031990 55.01 + 0.03 0.47 4.62e-02 53.28–55.01 1 Cook Yield 5 MARC0036560 66.10 + 0.03 0.31 3.30e-02 62.76–66.10 1 Loin Muscle Area 16-wk 6 ASGA0105067 69.37 + 0.04 0.29 3.16e-02 69.37–69.91 4f Last-Rib Backfat 13-wk 6 ALGA0104402 147.74 – 0.07 0.43 2.33e-05 108.75–151.35 6f Carcass 10th-Rib Backfat 6 M1GA0008917 144.60 – 0.12 0.45 8.11e-08 108.75–153.26 6f Last-Rib Backfat 10-wk 6 ALGA0104402 147.74 – 0.07 0.35 5.25e-05 114.92–153.13 6f Loin Weight 6 ASGA0029651 144.64 – 0.06 0.30 2.24e-03 135.37–150.06 4f Last-Rib Backfat 22-wk 6 ALGA0104402 147.74 – 0.07 0.49 8.80e-05 135.85–150.04 4f 10th-Rib Backfat 13-wk 6 ALGA0104402 147.74 – 0.05 0.43 3.05e-03 142.22–150.04 4 Last-Rib Backfat 16-wk 6 ALGA0104402 147.74 – 0.06 0.47 2.63e-04 142.75–153.05 5 Last Lumbar Backfat 6 ALGA0104402 147.74 – 0.05 0.41 1.38e-02 142.75–153.13 5 10th-Rib Backfat 10-wk 6 ASGA0029651 144.64 + 0.06 0.41 1.56e-02 144.55–147.74 2 10th-Rib Backfat 22-wk 6 ALGA0104402 147.744 – 0.04 0.47 2.69e-02 144.60–147.74 2 10th-Rib Backfat 16-wk 6 ALGA0104402 147.74 – 0.06 0.49 2.14e-03 0 2 10th-Rib Backfat 19-wk 6 ALGA0104402 147.74 – 0.05 0.50 8.72e-03 147.74–147.74 2 Last-Rib Backfat 19-wk 6 ALGA0104402 147.74 – 0.06 0.57 3.41e-04 147.74–150.04 2 Number of Ribs 7 ALGA0043983 98.51 + 0.12 0.36 4.19e-09 59.41–111.38 10 Cook Yield 8 DRGA0008986 134.93 – 0.03 0.31 2.01e-02 134.93–134.93 1 Dressing Percent 11 M1GA0014839 6.88 + 0.03 0.24 4.33e-02 1.72–6.88 2 Loin Weight 11 ALGA0060368 4.51 – 0.03 0.30 3.10e-02 4.51–4.51 2f Last-Rib Backfat 10-wk 12 ASGA0054658 41.73 – 0.02 0.35 4.03e-02 41.73–41.73 2 Protein Percent 15 MARC0093624 122.22 + 0.21 0.38 8.71e-20 48.78–131.28 21f 24-h pH 15 MARC0093624 122.22 + 0.09 0.19 3.35e-07 110.22–131.45 16f Cook Yield 15 MARC0093624 122.22 + 0.15 0.31 3.61e-13 113.26–132.44 16f Drip Loss 15 MARC0093624 122.22 – 0.13 0.28 4.20e-11 113.26–137.87 17f SP Overall Tenderness 15 H3GA0052416 121.81 + 0.07 0.28 3.59e-05 118.55–122.90 16f WBS 15 MARC0093624 122.22 + 0.06 0.26 4.07e-04 120.03–122.90 16f SP Juiciness 15 H3GA0052416 121.81 + 0.04 0.07 5.06e-03 120.03–122.22 16f SP Tenderness 15 H3GA0052416 121.81 + 0.07 0.29 1.90e-05 120.03–122.90 16f SNP significantly associated with phenotype aEffect of peak pQTL SNP on phenotype, positive indicates B allele increases phenotypic trait bProportion of phenotypic variance explained by peak SNP cGWAS qvalue for peak SNP dInterval for pQTL: start and end position eNumber of eQTL co-localized with the pQTL fContains at least one eQTL significantly associated with the phenotype transcript expression variance (Fig. 4). A second pQTL for variance and 10.7% of the gene expression variance. Re- loin weight was mapped on SSC11 and was significantly duced loin weight was associated with reduced expression associated with a novel transcript (SSC11:2.19), which co- of the SSC11:2.19 transcript (Fig. 4). incides with the uncharacterized locus LOC110255792. Considering the pQTL for meat quality and carcass The peak pQTL marker for loin weight on SSC11 composition traits with their associated eQTL reveals two (ALGA0060368) accounted for 2.7% of the phenotypic genomic regions of particular note. A 7.90-Mb region on Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 8 of 19

Fig. 4 Proportion of variance explained by peak pQTL SNP for phenotypes (blue) and gene transcript abundance (green). Traits are shown on the x-axis, and the proportion of phenotypic variance explained by the SNP marker is shown on the y-axis. Directionality of bar plots indicates SNP effect on phenotype or gene expression (i.e., increase or decrease)

SSC2:4.341–12.242-Mb was associated with the FERM correlation 0.89). The expression of eight (CEP128, domain-containing 8 gene (FRMD8) and WBS, sensory CHRNA9, MRLN, NQO1, PFKFB3, PKP2, SUCLG2, panel tenderness and overall tenderness phenotypes, and TEX9) of the eleven genes associated with the a 110.21 Mb region on SSC15:27.666–137.874-Mb was as- H3GA0052416 marker were significantly correlated with sociated with 11 genes and eight meat quality or carcass the protein percent phenotype (r = − 0.34 ± 0.05, FDR ≤ composition phenotypes (Tables 3 and 4). Significant 0.05; Table S4 in Additional file 1). These results suggest negative correlations were observed between WBS and all a potential candidate variant(s) on SSC15 accounting for three sensory panel phenotypes as expected for these traits a significant portion of phenotypic variation for meat (r = − 0.44 ± 0.14, p-value ≤8e-05, Fig. 4); more force quality and carcass composition phenotypes, as well as needed to break myofibers (i.e., higher shear force values) individual gene expression variation. Since the two was correlated with lower meat tenderness based on sub- markers (H3GA0052416 and MARC0093624) are in jective scores evaluated by a trained sensory panel. The high LD, the proportion of phenotypic and gene expres- peak pQTL markers, M1GA0002229 and H3GA0005676, sion variance was estimated for the H3GA0052416 for meat quality traits on SSC2 accounted for approxi- marker for all eight phenotypes and eleven gene tran- mately 5 % of the phenotypic variance and 8 % of FRMD8 scripts including CCDC60, CEP128, CHRNA9, CIT, gene expression variance (Fig. 4) with increased expres- MRLN, NQO1, PFKFB3, PKP2, SUCLG2, TEX9, and a sion of FRMD8 associated with increased sensory panel novel transcript SSC15:48.94-Mb (mapped to the unchar- tenderness and overall tenderness scores and decreased acterized locus LOC110257028). The H3GA0052416 WBS. High linkage disequilibrium (LD) was observed marker accounted for 4–16% of the phenotypic variance between these two SNPs (r =0.64). and approximately 23% of the gene expression variance Eleven of the eQTL significantly associated with pQTL (Fig. 4). The B allele of the H3GA0052416 marker was as- were distant regulators of gene expression, and all of sociated with increased expression of the eleven genes, these were also associated with a plausible hotspot and was also associated with an increase in sensory panel within the 110.21-Mb region on SSC15. The SSC15 scores and drip loss, and a decrease in WBS, 24-h pH, plausible hotspot marker H3GA0052416 was the peak cook yield and protein percent (Fig. 4). pQTL marker for sensory panel juiciness, tenderness The gene protein kinase AMP-activated non-catalytic and overall tenderness (Table 3), as well as the peak subunit gamma 3 (PRKAG3) maps to this region of eQTL marker for seven gene transcripts (Table 4). The SSC15, and variants of PRKAG3 have been implicated in peak pQTL marker for WBS, 24-h pH, cook yield, drip affecting meat quality phenotypes [22, 23]. We geno- loss and protein percent on SSC15 (MARC0093624) is typed all F2 animals for two PRKAG3 coding SNPs [21] in high LD with the H3GA0052416 marker (Pearson and included these SNP in our GWAS. However, the Velez-Irizarry ta.BCGenomics BMC al. et

Table 4 Expression QTL significantly associated with phenotypic traits Gene SSC Gene h2 SSC eQTL Interval eQTL Regulatorb Peak eQTL SNP Position Ec valued Phenotypee

TEX9 1 3.76e-01 15 27.666–122.219 Distant H 3GA 0052416 121.806 + 6.09e-04 M eat Quality, Protein Percent (2019)20:3 FRM D8 2 4.00e01 2 6.383–12.242 Local ASGA0098756 12.242 – 7.25e-04 WBS, SP Tenderness, SP Overall Tenderness PK P2 5 2.91e-01 15 121.806–121.873 Distant H 3GA 0052416 121.806 – 9.47e-03 M eat Quality, Protein Percent NQO1 6 3.28e-01 15 121.806–122.219 Distant H 3GA 0052416 121.806 + 1.74e-04 M eat Quality, Protein Percent H PN 6 3.62e-01 6 43.809–72.625 Local ALGA0113420 68.085 + 1.20e-03 Loin MuscleArea16-wk Last-Rib Backf at 10, 13, 22 M ETTL4 6 1.70e-01 6 99.932–136.022 Local MA RC0107785 101.29 – 9.55e-04 wk, Carcass 10th- Ri b Backf at, Loin Weight SSX 21 P 6 4.60e-01 6 120.497–143.338 Local ALGA0104761 141.317 + 5.27e-03 Carcass 10th- Rib Backf at, Loin Weight CEP128 7 4.47e-01 15 121.562–122.219 Distant H 3GA 0052416 121.806 + 1.39e-04 M eat Quality, Protein Percent CH RNA 9 8 4.21 e-01 15 120.031–122.219 Distant D I A S0000678 121.562 – 7.96e-05 M eat Quality, Protein Percent PFKFB3 10 7.44e-03 15 121.806–121.873 Distant M A RC0027291 121.873 – 5.03e-03 M eat Quality, Protein Percent SSC11:2.19a 11 5.23e-01 11 0.215–5.406 Local ALGA0060277 2.798 1.05e-04 Loin Weight SUCLG2 13 2.34e-01 15 120.031–121.873 Distant M A RC0027291 121.873 – 1.06e-03 M eat Quality, Protein Percent CI T 14 1.90e-01 15 121.562–122.219 Distant H 3GA 0052416 121.806 + 9.14e-06 M eat Quality, Protein Percent CCDC60 14 3.90e-01 15 121.806–122.767 Distant H 3GA 0052416 121.806 + 1.67e-03 M eat Quality, Protein Percent M RLN 14 4.00e-01 15 121.562–122.219 Distant M A RC0027291 121.873 – 1.54e-04 M eat Quality, Protein Percent SSC15:48.94a 15 1.75e-01 15 121.562–121.873 Distant SC H 3GA 0052416 121.806 + 4.67e-03 M eat Quality, Protein Percent aNovel gene transcripts: Sus scrofa chromosome and start position bRegulator type for eQTL cEffect the peak eQTL marker has on the gene’s expression: positive indicates B allele increases and negative indicates B allele decreases expression dqvalue of peak eQTL marker (FDR < 0.01) ePhenotypes significantly associated with the gene’s expression. Meat Quality includes the phenotypes for sensory panel juiciness, tenderness and overall tenderness, Warner Bratzler Shear Force, Cook Yield, Drip Loss and 24-h pH ae9o 19 of 9 Page Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 10 of 19

eQTL scan did not reveal associations with either of the define cis-acting (i.e., local) versus trans-acting (i.e., PRKAG3 markers. To further assess the effect of distant) regulation. For instance, distance thresholds be- PRKAG3, we performed a conditional analysis to esti- tween 1-Mb and 10-Mb have been used in recent pig mate the significance of these markers on identified eQTL maps [10, 13, 28–31]. Human eQTL scans have eQTL (Additional file 1 Table S5). Only one gene, used more conservative distance thresholds of 100-Kb – NQO1, was significantly associated with the PRKAG3 500-Kb between gene position and eQTL to declare T30 N SNP (FDR ≤ 0.01), where T30 N accounted for up local regulation [32, 33]. A shorter local threshold is lo- to 12% of the gene expression variance (data not shown). gical for human eQTL studies because they typically Given the high signal of the H3GA0052416 marker on show higher resolution due to increased SNP density SSC15 for various genes and meat quality traits, we esti- (millions of genotyped markers [32]), and the extent of mated the proportion of phenotypic variance explained by LD is less than in livestock populations due to greater both H3GA0052416 and the PRKAG3 T30 N marker for genetic diversity in human populations [33]. In this meat quality and carcass composition traits (Additional study, we present an alternative to the use of a fixed dis- file 2 Figure S4). The PRKAG3 T30 N marker accounted tance for declaring local versus distant eQTL effects. for 0.1–2% of phenotypic variance for meat quality traits, This is important because the range of a mapped eQTL whereas the H3GA0052416 marker accounted for 2–14%. will depend on the LD pattern at the QTL genomic pos- This analysis shows the H3GA0052416 marker accounts ition. Building upon previous approaches to determine for a greater proportion of phenotypic variance than the local regulation [11, 15, 34, 35] in eQTL linkage maps, PRKAG3 T30 N SNP. this study considered the significance of each individual marker surrounding the plausible position range of the RT-qPCR confirmation of CHRNA9 eQTL interval to distinguish between local and distant The statistical model identified 24 eQTL mapped to a modes of action. In cases where there are no genotyped 125-Mb region on SSC15. Eleven of these eQTL markers between the plausible position of the eQTL co-localized with pQTL for meat quality and carcass interval and the location of the associated gene, there is composition traits, and among these the CHRNA9 gene not sufficient information to determine local versus dis- was selected for verification using RT-qPCR (Fig. 4). tant; here we consider this scenario as plausible local CHRNA9 is implicated in catecholamine secretion and regulation. We note that in our study the median dis- the adaptive response to chronic stress [24], and is es- tance between plausible local eQTL regulators and their sential for muscle contraction [25]. The genomic pos- associated gene was 31-Kb, which is a shorter distance ition of the CHRNA9 gene is on SSC8: 31.44–31.51-Mb, than eQTL designated as local for other pig eQTL map- and the eQTL associated with this gene mapped to ping studies [10, 13, 28–31]. Therefore, it is feasible that SSC15, therefore exhibiting distant-acting regulation of most of these regulators may be acting locally, since CHRNA9 gene expression. RT-qPCR was performed to cis-acting transcription factor binding sites have been confirm the expression pattern of the CHRNA9 gene in found located ~ 100 kb from the mapped position of a longissimus dorsi muscle. Pearson correlation between gene transcript [36]. However, without a denser SNP set the ΔCt and RNA-seq log-cpm for CHRNA9 transcript and/or a larger population size, we cannot definitively abundance was − 0.58. The marker DIAS0000678 was identify the mode of action of these eQTL. A potential significantly associated with both RNA-seq and ΔCt for way to further investigate if these eQTL are acting CHRNA9 (p-value ≤4.23e-06), exhibiting a significant locally or distantly would be through allele-specific ex- dominant effect with the B allele associated with in- pression analyses [14]. creased CHRNA9 transcript abundance (p-value ≤0.05, Heritability of gene expression contributes to our under- Additional file 1 Table S6). standing of the inheritance of gene expression regulation. Estimatingtheheritabilityofgeneexpressioniscommonin Discussion human eQTL studies to elucidate the genetic contribution For this study, we identified eQTL for longissimus dorsi of gene expression variation and its influence on the diver- muscle transcripts from pigs in an F2 resource popula- gence of complex traits [32, 33, 37, 38]. Human studies tion, and we declared local versus distant eQTL effects have shown higher heritability estimates for housekeeping based on LD structure. When an eQTL and the associ- genes and genes with local eQTL, whereas genes with dis- ated gene are located on the same chromosome, the low tant eQTL tend to exhibit lower heritability [32, 37, 38]. resolution of the swine genome due to long range LD Bryois et al. [37] suggested a fraction of missing heritability [26, 27] limits the ability to distinguish between may be due to common variants with both local and distant cis-acting and trans-acting eQTL. Most eQTL associ- effects on gene expression, with the latter being of small ation studies use a fixed distance threshold between the effect size. Examples of local eQTL with large distant effects position of the eQTL interval and the gene transcript to in human studies include variants influencing the Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 11 of 19

expression of transcription factor genes or histone methyl- composite population [42], reported QTL for slice shear transferase genes [37]. Heritability of gene expression has force (a technique similar to WBS) in the same genomic re- not been emphasized in pig eQTL studies, except for one gion as this study. Zhang et al., [41] identified FRMD8 to report where heritability was used as a filtering criteria to beoneoffourgenesintheregiontoplayaroleinporkten- prioritize genes [35]. In this study, we estimated derization, and the peak SNP reported by Nonneman et al. narrow-sense heritability for all gene expression profiles [42] was the same peak SNP identified in our analysis and determined significance with likelihood ratio tests. (H3GA0005672). We showed with our conditional analysis Consistent with previous studies in humans, the observed that increased expression of FRMD8 was associated with heritabilities for genes with distant eQTL were significantly improvements in pork tenderness. FRMD8 is a member of lower than for locally regulated genes [32]. This trend is the FERM (Four-point-one, Ezrin, Radixin, Meosin) protein consistent with previous findings where genes influenced superfamily known to possess both structural and signaling by many distant factors of small effect tend to exhibit lower functions including numerous protein-binding interactions heritability than genes with local regulation. Testing for sig- mainly in the cytoskeleton of cells [43]. This includes inter- nificant additive genetic effects of transcript abundance in actions with transmembrane ion channels and membrane outbred animal populations requires a large sample size to lipids including the phosphatidylinositol 4,5-bisphosphate increase power to detect smaller effects. In our GWA scan, (PIP2). PIP2 is the precursor of inositol 1,4,5-triphosphate 2+ we were able to capture the variance associated with gene (IP3)involvedinCa signaling [44–46]andIP3 has been transcripts subject to genetic control with low heritability. suggested as a potential indicator of meat tenderness in 2+ A previous eQTL scan performed with 57 muscle tissue beef cattle [47]. The activation of the PIP2 Ca signaling samplesfromanF2swinepopulationobservedanaverage system controls diverse cellular processes in numerous tis- heritability of 0.45 for eQTL genes [35]. While this value is sues [48]. In skeletal muscle the sarcoplasmic reticulum greater than the average heritability observed in our study ryanodine receptor is the Ca2+ release channel, however, (0.32), Liaubet et al. [35] limited the eQTL scan to gene PIP2 has been localized to the transverse tubular mem- transcripts with heritability greater than 0.05. The use of a brane, and IP3 receptors have been found in differentiated heritability threshold to filter genes in eQTL studies may muscle fibers and implicated in excitation-contraction miss potential associations, especially those of low effect coupling (for review see Csernoch et al. [49]). Thus, such as distant eQTL, which we show to have lower aver- FRMD8 may play a role in Ca2+ signaling and age heritability estimates. excitation-contraction coupling of skeletal muscles through We identified three gene transcripts that were associ- interactions with PIP2. ated with pQTL for fat deposition and carcass compos- Similar to FRMD8, the MRLN gene is also implicated ition traits on SSC6. One of these eQTL genes, synovial in muscle contraction. MRLN encodes myoregulin, a sarcoma X breakpoint 2 interacting protein (SSX2IP), micropeptide inhibitor of the sarco/endoplasmic was significantly associated with pQTL for carcass 10th reticulum Ca+2 ATPase (SERCA). SERCA regulates re- rib backfat and loin weight. An eQTL was previously laxation after muscle contraction, specifically by pump- identified for this gene on SSC6 using microarray data ing Ca+2 back to the sarcoplasmic reticulum. Binding from the same animals used in this study, and consistent myoregulin to SERCA lowers its affinity to Ca+2, redu- with our results, Peñagaricano et al. [39] reported a cing the rate of Ca+2 reuptake into the sarcoplasmic negative causal effect of increased expression of SSX2IP reticulum [50]. Increased expression of MRLN was asso- on backfat thickness [39]. In addition, SSX2IP has been ciated with improvements in pork tenderization, de- associated with waist to hip ratio, a common measure of creased 24-h pH and increased drip loss in our study. body fat distribution, in women of African descent [40]. The observed effect of MRLN gene expression on meat The genes associated with pQTL for tenderness phe- quality phenotypes may be due to its involvement in notypes on SSC2 or meat quality phenotypes on SSC15 regulating muscle contractility and calcium signaling, share biological processes known to directly influence which have a direct effect on postmortem proteolysis. the organoleptic properties of meat, including calcium Additional genes implicated in calcium signaling and signaling (FRMD8, MRLN, PKP2 and CHRNA9), energy associated with meat quality phenotypes and the metabolism (SUCLG2 and PFKFB3), redox hemostasis H3GA0052416 marker were the PKP2 and CHRNA9 (NQO1 and CEP128) and cytoskeletal structure (CIT and genes. PKP2 encodes a plakophilin protein known to CCDC60). One of the genes related to calcium signaling localize to cell desmosomes and nuclei, and play a role is the FERM domain containing 8 (FRMD8)geneassoci- in linking cadherins to intermediate filaments in the ated with pQTL for WBS, and sensory panel tenderness cytoskeleton. In mouse cardiac muscle, PKP2 has been and overall tenderness on SSC2. Two independent GWAS, shown to regulate the transcription of genes controlling one in a crossbred commercial pig population [41]andan- intercellular calcium homeostasis, and reduced expres- other in a multigenerational Landrace-Duroc-Yorkshire sion of PKP2 decreases the expression of several calcium Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 12 of 19

signaling genes including the cardiac muscle ryanodine However, in our study, increased expression of PRKFB3 receptor [51]. In this study, increased expression of was associated with increased pork tenderness. Thus, PKP2 was associated with improvements in pork tender- similarly to PRKAG3, PRKFB3 may be involved in post- ness, and decreases in 24-h pH, protein percent and mortem glycolytic potential. cook yield, suggesting a role for this gene in modulating The CEP128 gene is related to the PI3K-Akt-mTOR sig- skeletal muscle calcium signaling during the conversion naling pathway. Centrosomal protein 128 (CEP128)ispart of muscle to meat. The CHRNA9 gene is one of sixteen of the centrosomal protein family, including CEP55 that subunits of the nicotinic acetylcholine receptor (AChR). has been implicated in cancer progression [54]. Mutations These ligand-gated ion channels permit the transmission within CEP128 have been associated with an aggressive of presynaptic acetylcholine release and postsynaptic ex- type of lymphoma, the diffuse large B-cell lymphoma citatory potential. Found only in neuronal tissue, (DLBCL) [55]. Functional gene studies have not been per- CHRNA9 is one of three AChR containing only α sub- formed for CEP128, however mutations identified in re- units [25](α9-AChR), and in neuromuscular junctions fractory DLBCL patients, including those in CEP128,were AChR are essential for muscle contraction [25]. Since associated with PI3K-Akt-mTOR signaling pathways and α9-AChR possess higher calcium permeability, they play increased mitochondrial oxidative phosphorylation, and a role in catecholamine secretion and the adaptive re- play an important role in treatment resistance [55]. The sponse to chronic stress [24]. In this study, increased ex- PI3K-Akt-mTOR pathway is upregulated in cancer cells, pression of CHRNA9 was associated with improved controlling the survival and proliferation of these cells. In tenderness scores, increased drip loss, and decreased our study, increased expression of CEP128 was associated cook yield, protein percent and 24-h pH. In addition, we with improved tenderness scores, potentially involving verified the expression of CHRNA9 in skeletal muscle PI3K/Akt/mTOR signaling. with RT-qPCR and confirmed a significant dominance The Edomucin (EMCN) gene associated with a effect of the peak eQTL SNP (DIAS0000678) on local-acting eQTL on SSC8 plays a critical role in angio- CHRNA9 gene expression. Changes in the expression of genesis. Angiogenesis is the process of new blood vessel CHRNA9 may potentially regulate the postsynaptic exci- formation with its key regulator, vascular endothelial tatory potential during the conversion of muscle to meat growth factor (VEGF), triggering downstream signaling thereby influencing Ca2+ release to the cytoplasm, apop- cascades including MAPK-ERK1/2, PI3k/Akt and totic mitochondrial changes and proteolytic enzymatic p38-MAPK pathways [56]. These signaling pathways activity. promote endothelial cell migration, proliferation, and Additional genes associated with meat quality traits on survival and are activated by HIF-1 which induces VEGF SSC15 (PFKFB3, CEP128, NQO1 and SUCLG2) were im- expression [57]. While this eQTL is not directly associ- plicated in biological processes related to redox homeo- ated with a phenotype in our population, it is connected stasis and energy metabolism. The PFKFB3 gene to the pathways regulated by the genes associated with regulates the synthesis and degradation of fructose-2, the H3GA0052416 marker on SSC15. 6-bisphosphate and fructose-6-phosphate in the process The remaining two genes, NQO1 and SUCLG2, were of glucose metabolism. The promoter of the PRKFB3 associated with improvements in meat tenderization and gene contains hypoxia-inducible factor-1 (HIF-1) bind- pH decline. The nuclear erythroid-2-p45-related factor-2 ing sites [52]. The transcription factor HIF-1 is a master (Nrf2) is a transcription factor known to regulate redox regulator of oxygen homeostasis by activating several homeostasis and anti-inflammatory response by control- downstream pathways including the mitogen-activated ling the expression of Phase I and Phase II anti-oxidant protein kinase (MAPK), mammalian target of rapamycin enzymes containing the antioxidant response element (mTOR), phosphoinositide 3-kinase-protein kinase B (ARE; cis-acting regulatory or enhancer sequence) in (PI3K-Akt), vascular endothelial growth factor (VEGF) their promoter regions. NQO1 (NADPH quinone and calcium signaling pathways, as well as anaerobic oxidoreductase-1) is one of these enzymes whose expres- metabolism. PFKFB3 is consistently overexpressed in sion is induced by Nrf2 in several tissues [58–61]. many tumor cells, and knockdown of PFKFB3 promotes Consequently, knockdown of Nrf2 has been reported to apoptosis of tumor cells [52]. Rapidly proliferating tumor significantly decrease expression of NQO1 in both cells have the ability to increase glucose uptake by using mouse skeletal muscle [60] and C2C12 mouse myotubes anaerobic glycolysis as the primary source of energy, [61]. In early postmortem muscle, the antioxidant known as the Warburg effect. Taken together, PFKFB3 is defense system is speculated to influence proteolysis and critical for cell proliferation and survival by regulating glu- thereby meat tenderization [5]. Increased expression of cose metabolism and prevents apoptosis through the acti- NQO1 in this study was associated with several meat vation of cyclin-dependent kinases [52, 53]. No reports quality traits including tenderness, pH and drip loss phe- have suggested a role for PRKFB3 in meat quality. notypes implying a significant role in post-mortem Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 13 of 19

proteolysis. The succinate-CoA ligase GDP-forming beta other ten gene transcripts mapped to different chromo- subunit (SUCLG2) has been implicated in the SUCL- somes than their associated eQTL. These results illustrate G1-related mitochondrial DNA depletion syndrome af- the advantage of the joint analysis of gene expression pro- fecting brain and skeletal muscle tissues. Individuals files and trait phenotypes to uncover the genetic architec- affected by this syndrome present an array of symptoms ture of polygenic traits. In this study, increased expression including spasmodic muscle contractions, contracture or of the 11 genes was associated with improvements in meat destruction of muscle cells and hypoglycemia [62]. quality phenotypes. Moreover, this QTL region harbors a Knockdown of the SUCLG2 gene in fibroblasts was re- plausible putative hotspot (H3GA0052416) regulating the ported to decrease mitochondrial DNA, mitochondrial expression of all 11 gene transcripts. Breitling et al. reported nucleoside diphosphate kinase and cytochrome c oxidase the high false positive rate associated with hotspot discov- activities [63]. These results highlight the critical role ery. Therefore, to mitigate this, we used a higher threshold SUCLG2 plays in mitochondrial DNA maintenance and of significance to detect eQTL. The H3GA0052416 marker ATP production. In our study, increased expression of on SSC15 was also associated with multiple meat quality SUCLG2 was associated with improvements in meat phenotypes. The high correlation observed between the 11 quality traits suggesting a potential role in regulating gene expressions, and between the eight meat quality ATP production and postmortem pH decline. phenotypes suggests the potential that these associations In addition to genes involved in specific biological are due to a master regulator on SSC15 (i.e., a putative functions, genes encoding structural proteins were also hotspot). The PRKAG3 gene has been suggested as such a observed to be associated with the H3GA0052416 regulator of meat quality traits in pigs. PRKAG3 regulates marker on SSC15 (CIT and CCDC60). CIT, citron glycogen potential, which has a cascading effect in post- Rho-interacting serine/threonine kinase, is considered to mortem metabolism. The SNP panel used in this study be a scaffold protein that binds to several mitotic pro- does not have sufficient coverage of the PRKAG3 gene. To teins, and knockout of CIT leads to cytokinetic defects. address this, our F2 population was genotyped for two One such protein-protein interaction involves the known PRKAG3 SNPs [21]. However, PRKAG3 did not two-pore channel 1 (TPC1), which Horton et al. [64] re- explain the relationship observed in the putative hotpot. A ported to cause disruption in myosin light chain phos- missense polymorphism within the PRKAG3 gene, T30 N phorylation (pMLC). In skeletal muscle, pMLC has been SSC15:120.865-Mb, was significantly associated with just associated with age-related muscle dysfunction [65], and one of the 11 genes, NQO1, despite showing significant decreased pMLC is associated with a reduced fraction of association with all eight meat quality phenotypes in this myosin heads interacting with thin filaments [65]. Thus, population [21]. increased expression of CIT could potentially increase muscle breakdown, which is consistent with our findings Conclusions where higher expression of CIT was associated with im- In summary, the joint analysis of pQTL with eQTL from provements in pork tenderization, and reduced protein our well-characterized pig resource population identified content and cook yield. CCDC60 is a coil-coil domain molecular markers significantly associated with both protein, which are believed to act as “cellular velcro” economically important phenotypes and gene transcript holding together molecules, cellular structures and tis- abundance. This approach revealed both local- and sues [66]. The biological function of CCDC60 is unknown, distant-acting regulators of gene expression influencing but recent GWAS have associated this gene with the meat quality, carcass composition and growth traits. neurological disorder schizophrenia in humans [67]. Prote- These phenotypic traits are correlated, and we show omic analysis of post-mortem pre-frontal cortex of schizo- how correlated phenotypes exhibit correlated gene ex- phrenia patients and non-schizophrenia individuals pression measured through a plausible hotspot con- identified differentially expressed proteins involved in cal- tained within QTL regions for both expression and cium homeostasis, cytoskeleton assembly and energy me- phenotypic traits. We highlight novel candidate genes tabolism [68]. Therefore, it is feasible that similar functions with specific roles in cytoskeletal structure and signaling may occur in skeletal muscle tissue. In this study, increased pathways regulating meat quality phenotypes including expression of CCDC60 was associated with tenderness, pH, redox hemostasis (NQO1 and CEP128), energy metabol- cook yield and drip loss phenotypes implicating the role of ism (SUCLG2 and PRKFB3), Ca2+ signaling (FRMD8, this gene in the conversion of muscle to meat. MRLN, PKP2 and CHRNA9) and cytoskeletal structure Eleven eQTL genes were enriched in pQTL for meat (CIT and CCDC60) during the initial conversion of quality traits on SSC15 (PFKFB3, SUCLG2, CIT, CCDC60, muscle to meat. Taken together the identified genes and MRLN, PKP2, NQO1, CEP128, CHRNA9, TEX9 and a novel their associated functions and pathways increase our transcript SSC15:48.94). The novel transcript mapped to an knowledge of the genomic architecture of meat quality uncharacterized locus, LOC110257028, on SSC15. The phenotypes. Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 14 of 19

Methods RNA extraction and RNA sequencing Pig population and phenotype collection Tissue samples were collected immediately post mortem Animal housing and care protocols were evaluated and from the longissimus dorsi muscle, flash frozen in liquid approved by the Michigan State University All University nitrogen and stored at − 80 °C until processed [17]. RNA Committee on Animal Use and Care (AUF # 09/03– extraction was performed with the miRNeasy Mini Kit 114-00). The experimental design, phenotyping and sample (Qiagen, Germantown, MD) following the manufac- collection for the MSUPRP has been reported previously turer’s protocol. Quality and quantity of extracted total [17–19]. All MSUPRP pigs were reared at the Michigan RNA were determined using the Agilent 2100 Bioanaly- State University Swine Teaching and Research Center, and zer (RIN ≥ 7). Sequencing was performed at the Mich- pigs for this study were euthanized by humane slaughter in igan State University Research Technology Support a USDA inspected abattoir at Michigan State University. Facility. Libraries for 24 samples were prepared using The MSUPRP was developed from 4 Duroc boars and 15 the Illumina TruSeq RNA Library Prep Kit v2, and se- Pietrain sows [18, 19]. From the F1 progeny, 56 animals (6 quenced on the Illumina HiSeq 2000 platform (2 × 100 males and 50 females) were retained to produce the F2 bp paired-end reads). The remaining 152 libraries were generation, which included 1259 animals from 142 litters. prepared using the Illumina TrueSeq Stranded mRNA A total of 67 phenotypic traits were collected for the F2 Kit, and sequenced on the Illumina HiSeq 2500 platform generation [18, 19]. A subset of the F2 pigs (168) were se- (2 × 125 bp, paired-end reads). Base calling was per- lected for this study using a selective profiling scheme formed with the Illumina Real Time Analysis v1.18.61 based on extremes in loin muscle area and backfat thick- software, and the Illumina Bc12fastq v1.8.4 was used for ness phenotypes within litter (44 litters) and sex [69]. Sum- conversion to FastQ format. A total of 96 sequence files mary statistics for the 67 phenotypic traits (29 growth (741Gb) consisting of ~ 63 million short-reads per li- traits, 20 carcass composition traits and 18 meat quality brary were obtained from the HiSeq 2000 platform, and traits) in the F2 population, and the subset of animals used 1218 sequence files (~ 2 Tb) of ~ 23 million short-reads for this study are shown in Additional file 1 Table S7. per library were obtained from the HiSeq 2500 platform. Eight samples were removed from further analysis due to low sequence quality, leaving a total of 168 samples Genotyping for subsequent analyses. Sequence data has been SNP genotypes for the MSUPRP were available from deposited in the NCBI Sequence Read Archive accession prior studies [70, 71]. Genotyping was performed by number PRJNA403969. Neogen Corporation - GeneSeek Operations (Lincoln Raw RNA sequence reads were first filtered for adapter NE) using the Illumina PorcineSNP60 BeadChip [72] for sequences using Trimmomatic [73] followed by quality the F0, F1 and ~ 1/3 of the F2 population (including all trimming using Condetri where the first six bases at the F2 pigs used for the eQTL analysis), and the GeneSeek 3′ end and low quality reads were filtered out, retaining Genomic Profiler for Porcine Low Density (GGP-Porcine reads with a minimum length of 75 bases (Figure S5 in LD) for the remaining F2 pigs [70, 71]. Missing geno- Additional file 2). The quality of each sequenced nucleo- types were imputed with an accuracy of 0.97 [70, 71]. tide was evaluated on adapter filtered and quality Monomorphic markers and non-autosomal markers trimmed RNA-seq reads using the FASTX toolkit [74], were eliminated from further analysis, as were those and a mean Phred quality score of 37.01 ± 0.99 was ob- showing divergence from Mendelian inheritance rules. tained. After adapter and quality filtering, RNA-seq An updated genomic map for SNPs on the Sscrofa11.1 reads were mapped to the reference genome assembly genome assembly was obtained from Neogen (Lincoln Sus scrofa 11.1 using the splice aware aligner Tophat2 NE). Additional filtering was performed to exclude [75]. Sample-specific transcriptomes were assembled markers with a minor allele frequency lower than 0.01 using Cufflinks and merged with the reference genome and to reduce the degree of correlation between adjacent to create a set of known and novel isoforms using Cuff- markers (i.e., if a pair of neighboring markers had a cor- merge [76]. A total of 28,033 full-length transfrags were relation of allelic dosage greater than 0.95, one of the identified (Figure S5 in Additional file 2). Alignment sta- two markers was eliminated; this filtering was performed tistics and base coverage were obtained with SAMtools only for the eQTL analysis). Filtering resulted in 23,162 [77]. Samples showed on average 92.4% of sequencing markers for the eQTL analysis and 43,130 markers for reads mapping to the reference genome, and 73.3% were the pQTL analysis. Two coding SNPs in the protein unique and properly paired with their complementary kinase AMP-activated non-catalytic subunit gamma 3 sequence (Figure S5 in Additional file 2). Total gene ex- (PRKAG3) gene, I199V and T30 N [22, 23], were also ge- pression abundance was quantified using unique and notyped in the MSUPRP as previously described in properly paired reads using HTseq [78]. Genes with total Casiro et al. [21]. count abundance less than 168 were removed from Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 15 of 19

further analysis to reduce the number of genes with low effect, a is a vector of random additive genetic effects expression, leaving 15,249 gene transcripts for eQTL and e is a vector of random residual errors. The additive  ð ; σ2Þ analysis (Figure S5 in Additional file 2). genetic effects are assumed a N 0 G a with the gen- ′ omic relationship matrix [82], G = ZZ . Z is a matrix of RNA-seq count normalization and transformation normalized SNP genotypes, with elements: Expressed gene counts were normalized using the M−2p trimmed mean of M-values (TMM) to reduce systematic Z ¼ pPffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð3Þ technical biases of sequenced transcripts [79]. TMM ðÞ2pðÞ1−p normalization has been shown to control false positive where, M is the matrix of SNP genotypes and p is a vector associations [80]. The normalized gene counts were then with the frequency of each reference allele. The error term transformed to follow an approximately Gaussian distri- is e  Nð0; σ2 diagðwdÞÞwith a variance inversely pro- bution by calculating the log counts per million e std portional to the variance coefficients, wd . These variance (log-cpm) as described in Law et. al. [81]. Briefly, a lin- std coefficients account for the heteroskedasticity across ear model was fit to obtain the expected log-cpm for genes with different expression. The heritability of gene each gene, E(y)=xβ, where y are the log-cpm, x is a vec- expressions were calculated by taking the ratio of the vari- tor of ones and β is a vector of estimated regression co- ance of the additive genetic effects to the total phenotypic efficients. The residual standard deviations for each gene 2 ¼ σ2=ðσ2 þ σ2Þ and their calculated average log-cpm were used to esti- variance, h a a e . mate the mean variance trend, w^ , by fitting a LOWESS Statistical significance of heritability was determined ¼ ½ ðθ^Þ− ðθb ފ; curve [81]. Variance coefficients were standardized to using a likelihood ratio test, LR 2 logL logL 0 keep similar scales for residual variance and additive comparing the likelihood of the model represented in variance: ^ Eq. 1 ðLðθÞÞ and the likelihood of a null model that does 1 b pffiffiffiffi not include the genetic additive effect ðLðθ0ÞÞ . Testing w^ 2 2 wd ¼ X ð1Þ the null hypothesis σ ¼ 0 is equivalent to testing h =0. std 1 1 a pffiffiffiffi The likelihood ratios were compared to a chi-squared ^ n w distribution with one degree of freedom, and the result- ing p-value divided by 2 to account for the asymptotic where, wdstd are the variance coefficients, n is the total number of animals, and w^ is the estimated mean vari- distribution of the likelihood ratios that tend to follow a ance trend. The normalized log-cpm were used as the mixture of chi-square distributions with different de- grees of freedom [83]. Multiple test corrections were response variable, y, and the variance coefficients, wdstd , were used to model heterogeneity of error variance in performed using a FDR of 0.01 [84]. Differences in herit- the eQTL scan. This approach accounts for the mean ability between local and distant eQTL were determined variance relationship of each gene expression instead of with Wilcoxon rank sum test [85]. assuming equal variance for all observations. Genome wide association ^, ð^Þ Heritability of phenotype and gene expression The SNP effects, g and their variances Var g were esti- A genomic best-linear unbiased prediction (GBLUP) mated as a linear transformation of the BLUP breeding a^ model [70, 71] was used to estimate the heritability of values, , from Eq. 2 [86, 87]. A test statistic for the as- each phenotype and gene expression by fitting the fol- sociation of each marker with each phenotype or gene lowing equation: expression measure is computed by standardizing the SNP effects: y ¼ Xb þ a þ e; ð2Þ g^ T ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð4Þ where, y is a vector with measurements of a phenotype VarðÞg^ for each animal when estimating phenotypic heritability, and a vector with normalized log-cpm gene expression The p-values associated with this T test statistic were when estimating the heritability of gene expression. X is calculated using the Gaussian cumulative distribution Φ an incidence matrix of fixed effects including sex and function, , as follows: additional covariates unique to each phenotype [20, 21], p−value ¼ 21½Š−ΦðÞjjT ; ð5Þ and includes the transcriptional profiling selection scheme (i.e., within litter and sex extremes for loin and subject to multiple test corrections per each gene muscle area or back fat thickness) when analyzing gene (FDR ≤ 0.01) [84]. If a gene had more than one expression. The vector b contains the estimated fixed expressed transcript, the FDR was computed for the Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 16 of 19

merged p-values of all transcripts expressed for the analysis that tested the effect of the most significant gene. marker associated with the pQTL on the co-localized It has been demonstrated [86, 87] that the T test sta- eQTL gene expression, as follows: tistics and p-values resulting from Eqs. (4 and 5 are ¼ þ þ þ ; ð Þ equivalent to those obtained from fitting a single marker y Xb ZSNPbSNP a e 6 model, specifically the Efficient Mixed-Model Associ- where, y is the expression of the co-localized eQTL gene. ation (EMMA) model [88]. The X,b, a and e were previously described in Eq. 2. ZSNP is a vector of standardized marker genotypes for Local and distant regulators the pQTL peak marker, co-localized with the eQTL Due to low SNP density and long-range LD in this pig gene, and bSNP is the estimated marker effect. Type I population, distinguishing local versus distant regulation error rate of 0.05 and Bonferroni p-value cutoff based of gene expression is difficult. We applied the following on the number of tests performed (p-value ≤5.952e-04) algorithm to classify putative eQTL as local or distant was used to determine SNP effect significance. We also ’ regulators of a gene s expression: considered the effect the peak pQTL marker had on the eQTL peak by performing a linear transformation of the 1) An eQTL was defined as any gene whose BLUP breeding values from Eq. 6 to estimate the indi- expression was associated with at least one marker vidual SNP effects, and tested their significance as de- ≤ surpassing the significance threshold (FDR 0.01). scribed in Eqs. (4 and 5. Multiple test corrections were 2) The plausible position range of each eQTL was performed using an FDR ≤ 0.01 [84]. If fitting the top defined by the position of the first significant pQTL marker completely eliminated the eQTL interval, marker at the beginning of the QTL and last the two QTL were considered to be significantly co- significant marker at the end of the QTL. If the localized. The proportion of variance explained by the eQTL had only one marker association, the position peak pQTL markers for each co-localized eQTL was es- of the marker was used. timated as described in Casiro et al. [21]. Briefly, the 3) Given the mapped position of the gene profile (start variance associated with the co-localized peak pQTL and end position of the transcript), there are several σ2 marker, SNP, was estimated as: possibilities a. The associated eQTL plausible position range σd2 ¼ 2 ðÞ; ð Þ SNP b var ZSNP 7 overlaps totally or partially: Local eQTL b. The associated eQTL is on a different where, b2 is the calculated peak pQTL marker effect chromosome: Distant eQTL from Eq. 6, and the proportion of gene expression vari- c. The associated eQTL is on the same ance accounted for by the co-localized pQTL peak SNP σd2 =ðσd2 þ σb2 þ σb2Þ chromosome but does not overlap: is SNP SNP a e . The estimated additive genetic ≥ σ2 σ2 i. There are non-significant SNP (FDR 0.01) variance, a , and error variance, e , are obtained after between the mapped position of the gene fitting Eq. 6. A conditional analysis for each eQTL inter- profile and its associated eQTL range: Dis- val, fitting the peak SNP, was also performed to identify tant eQTL same chromosome potential additional peaks within an eQTL. Pearson cor- ii. There are no SNP between gene and eQTL relations among phenotypes and gene expressions were range (including the filtered SNP due to calculated using residuals from Eq. 6, and significance high LD): Plausible Local was determined with a t test and FDR ≤ 0.05. LD was es- timated between the peak SNP for the pQTL and colo- The distance between an eQTL and the corresponding calized eQTL. Equations 6 and 7 were also used to gene location was estimated as the difference between estimate the proportion of gene expression variance ex- the peak SNP and the nearest position of the gene plained by the PRKAG3 T30 N SNP for all identified transcript. eQTL to uncover eQTL significantly associated with PRKAG3, and the proportion of phenotypic variance ex- Co-localization analysis plained for meat quality phenotypes with an associated The genomic positions of the mapped eQTL were pQTL on SSC15. co-localized with pQTL previously identified for the F2 MSUPRP for growth, carcass composition and meat RT-qPCR quality traits. An eQTL was considered co-localized if its To verify the expression of CHRNA9, 28 animals were QTL position overlapped the mapped location of a selected based on the genotypes of the peak eQTL SNP pQTL. The statistical significance of each co-localized (10 animals per genotype equally weighted by sex except eQTL with pQTL was determined through a conditional for the AA genotype that had only eight animals, four Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 17 of 19

per sex). Total RNA was extracted from the longissimus kinase AMP-activated non-catalytic subunit gamma3; QTL: Quantitative trait muscle samples as described above, and 2 μg was reverse locus; SNP: Single nucleotide polymorphism; SSC: Sus scrofa chromosome; SUCLG2: Succinate-COA ligase GDP-forming beta subunit; TMM: Trimmed transcribed using the High Capacity cDNA Reverse mean M-values; WBS: Warner Bratzler shear force Transcriptase Kit with RNase inhibitor (Applied Biosys- tems, Foster City, CA). A custom Taqman Gene Expres- Acknowledgements The authors wish to thank Scott Funkhouser and Ryan Corbett for their sion Assay (ThermoFisher Scientific, Waltham, MA) was constructive feedback, the MSU Institute for Cyber-Enabled Research High designed for CHRNA9 using pig RNA sequence to span Performance Computer Center for providing the computing environment to exons 4 and 5 (determined based on the structure of the analyze and store data, and the MSU Research Technology Support Facility human CHRNA9 gene, Accession No. AC118275). The for performing the RNA sequencing. GeNorm [89] algorithm was used to select two reference Funding genes, PPIA (ThermoFisher Scientific Assay No. This work was supported by Agriculture and Food Research Initiative – Ss03394781_g1) and SDHA (ThermoFisher Scientific competitive grant no. 2014 67015-21619 from the USDA National Institute of Food and Agriculture. DVI and KRD were supported by fellowships from the Assay No. Ss03376909_u1), with the highest Food and Agricultural Sciences National Needs Graduate and Postgraduate gene-stabilizing measure to normalize the expression of Fellowships Grants Program award no. 2012–38420-30199 from the USDA CHRNA9. RT-qPCR was performed in triplicate using National Institute of Food and Agriculture. The funders had no role in the study design, data collection, analysis, and interpretation, or in writing the 50 ng cDNA and TaqMan Gene Expression Master Mix manuscript. for a final volume of 20 μl. Assays were run on a StepO- nePlus Real-Time PCR System (Applied Biosystems). Availability of data and materials Sequence data has been deposited in the NCBI Sequence Read Archive with The cycling conditions were 52 °C for 2 min, 95 °C for BioProject Number PRJNA403969 (Accessions SRX3174410 to SRX3174577). 10 min followed by 50 cycles of 95 °C for 15 s and 60 °C for 1 s. ΔCt values were calculated as the mean differ- Authors’ contributions CWE, ROB and JPS conceived of the study and supervised the research. DVI ence between the geometric mean of the reference genes performed the data analyses and drafted the manuscript. SC contributed to and the target gene. To verify the RNA-seq results, the the statistical framework used for the study. KRD assisted with RNA effect of the peak eQTL marker for CHRNA9 was mea- extraction and participated in discussions regarding data analysis. CWE, NER and ROB contributed to tissue and phenotype collection, NER performed sured using Eq. 6 with the response variable being the RNA extraction and RT-qPCR, and oversaw sequencing of samples. All au- ΔCt transcript abundance. Analysis of variance with a thors read and approved the final manuscript. type I error rate of 0.05 was used to determine signifi- cant additive and dominance effects of the peak Ethics approval and consent to participate Animal housing and care protocols were evaluated and approved by the CHRNA9 eQTL SNP (DIAS0000678). Michigan State University All University Committee on Animal Use and Care (AUF # 09/03–114-00). Additional files Consent for publication Not applicable. Additional file 1: Table S1. Expression quantitative trait loci (eQTL) mapped for longissimus dorsi muscle transcripts from the MSUPRP (n =168). Competing interests Table S2. Phenotypic QTL identified in the F2 MSUPRP. Table S3. Results of The authors declare that they have no competing interests. conditional analysis for expression QTL co-localized with phenotypic QTL. Table S4. Pearson correlations of genes associated with the H3GA0052416 marker on SSC15 and protein percent. Table S5. Conditional analysis: Publisher’sNote PRKAG3 SNP effect on eQTL gene’sexpression.Table S6. Comparison of Springer Nature remains neutral with regard to jurisdictional claims in RT-qPCR and RNA-seq CHRNA9 gene expression association with SNP published maps and institutional affiliations. DIAS0000678. Table S7. Summary statistics for phenotypic traits for the MSUPRP and the subsample used in this study. (XLSX 200 kb) Author details 1Department of Animal Science, Michigan State University, East Lansing, MI Additional file 2: Figure S1. Manhattan plots illustrating classification of 2 different types of gene expression regulation based on eQTL position. 48824, USA. Department of Fisheries and Wildlife, Michigan State University, Figure S2. Heritability of transcript profiles. Figure S3. Pearson correlations East Lansing, MI 48824, USA. among phenotypic traits with an associated pQTL. Figure S4. Proportion of phenotypic variance explained by PRKAG3 and H3GA0052416 SNP for meat Received: 10 July 2018 Accepted: 18 December 2018 quality traits. Figure S5. RNA-seq pipeline. (PDF 2134 kb)

References Abbreviations 1. Dekkers JCM. Commercial application of marker- and gene-assisted CEP128: Centrosomal protein 128; CHRNA9: Cholinergic receptor nicotinic selection in livestock : strategies and lessons. J Anim Sci. 2004;82:E313–28. alpha 9 subunit; CIT : Citron Rho-interacting serine/threonine kinase; 2. Kadarmideen HN, von Rohr P, Janss LLG. From genetical genomics to eQTL: Expression quantitative trait locus; FDR: False discover rate; systems genetics: potential applications in quantitative genomics and FRMD8: FERM domain-containing 8 gene; GBLUP: Genomic best-linear un- animal breeding. Mamm Genome. 2006;17:548–64. https://doi.org/10.1007/ biased prediction; GWAS: Genome wide association study; h2: Heritability; s00335-005-0169-x. LD: Linkage disequilibrium; MRLN: Myoregulin; MSUPRP: Michigan State 3. Goddard ME, Hayes BJ. Mapping genes for complex traits in domestic University Pig Resource Population; NQO1: NADPH quinone oxidoreductase animals and their use in breeding programmes. Nat Rev Genet. 2009;10: − 1; PFKFB3: 6-phosphofructo-2-kinase / fructose-2-,6-biphosphatase 3; PI3K- 381–91. https://doi.org/10.1038/nrg2575. Akt-mTOR: Phosphatidylinositol-3-kinase/Akt/mammalian target of rapamycin; 4. Van Der HAM S, GFW P, Plastow GS. Application of genomics to the pork PKP2: Plakophilin; pQTL: Phenotypic quantitative trait locus; PRKAG3: Protein industry. J Anim Sci. 2005;83(13):E1–8. Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 18 of 19

5. England EM, Schef TL, Kasten SC, Matarneh SK, Gerrard DE. Exploring the 26. Wimmers K, Murani E, Te Pas MFW, Chang KC, Davoli R, Merks JWM, et al. unknowns involved in the transformation of muscle to meat. Meat Sci. Associations of functional candidate genes derived from gene-expression 2013;95:837–43. profiles of prenatal porcine muscle tissue with meat quality and muscle 6. Hu Z, Park CA, Reecy JM, Animal T, Database QTL. Developmental progress deposition. Anim Genet. 2007;38:474–84. and current status of the Animal QTLdb. Nuc. 2016;44:D827–33. 27. Badke YM, Bates RO, Ernst CW, Schwab C, Steibel JP. Estimation of linkage 7. Wan X, Wang D, Xiong Q, Xiang H, Li H, Wang H. Elucidating a molecular disequilibrium in four US pig breeds. BMC Genomics. 2012;13:24. https://doi. mechanism that the deterioration of porcine meat quality responds to org/10.1186/1471-2164-13-24. increased cortisol based on transcriptome sequencing. Sci Rep. 2016;6: 28. Chen C, Wei R, Qiao R, Ren J, Yang H, Liu C. A genome-wide investigation 36589. https://doi.org/10.1038/srep36589. of expression characteristics of natural antisense transcripts in liver and 8. Rocha LM, Velarde A, Dalmau A, Saucier L, Faucitano L. Can the monitoring muscle samples of pigs. PLoS One. 2012;7:e52433. of animal welfare parameters predict pork meat quality variation through 29. Kogelman LJA, Zhernakova DV, Westra H, Cirera S, Fredholm M. An the supply chain ( from farm to slaughter )? J Anim Sci. 2016;94:359–76. integrative systems genetics approach reveals potential causal genes and 9. Hao Y, Feng Y, Yang P, Cui Y, Liu J, Yang C. Transcriptome analysis reveals pathways related to obesity. Genome Med. 2015;7:105. https://doi.org/10. that constant heat stress modifies the metabolism and structure of the 1186/s13073-015-0229-0. porcine longissimus dorsi skeletal muscle. Mol Gen Genomics. 2016;291: 30. Ponsuksili S, Murani E, Brand B, Schwerin M, Wimmers K. Integrating 2101–15. expression profiling and whole-genome association for dissection of fat 10. Ponsuksili S, Du Y, Murani E, Schwerin M, Wimmers K. Elucidating molecular traits in a porcine model. J Lipid Res. 2011;52:668–78. https://doi.org/10. networks that either affect or respond to plasma cortisol concentration in 1194/jlr.M013342. target tissues of liver and muscle. Genetics. 2012;192:1109–22. 31. Ponsuksili S, Murani E, Trakooljul N, Schwerin M, Wimmers K. Discovery of 11. Ponsuksili S, Jonas E, Murani E, Phatsara C, Srikanchai T, Walz C, et al. Trait candidate genes for muscle traits based on GWAS supported by eQTL- correlated expression combined with expression QTL analysis reveals analysis. Int J Biol Sci. 2014;10:327–37. biological pathways and candidate genes affecting water holding capacity 32. Yang S, Liu Y, Jiang N, Chen J, Leach L, Luo Z, et al. Genome-wide eQTLs of muscle. BMC Genomics. 2008;9:367. and heritability for gene expression traits in unrelated individuals. BMC 12. Wimmers K, Murani E, Ponsuksili S. Functional genomics and genetical Genomics. 2014;15:13. https://doi.org/10.1186/1471-2164-15-13. genomics approaches towards elucidating networks of genes affecting 33. DixonAL,LiangL,MoffattMF,ChenW,HeathS,WongKC,etal.A meat performance in pigs. Brief Funct Genomics. 2010;9:251–8. genome-wide association study of global gene expression. Nat Genet. 13. Ma J, Yang J, Zhou L, Ren J, Liu X, Zhang H, et al. A splice mutation in the 2007;39:1202–7. PHKG1 gene causes high glycogen content and low meat quality in pig 34. Cánovas A, Pena RN, Gallardo D, Ramirez O, Amills M, Quintanilla R. skeletal muscle. PLoS Genet. 2014;10:e1004710. Segregation of regulatory polymorphisms with effects on the gluteus 14. Ernst CW, Steibel JP. Molecular advances in QTL discovery and application medius transcriptome in a purebred pig population. PLoS One. 2012;7: in pig breeding. Trends Genet. 2013;29:215–24. e35583. https://doi.org/10.1371/journal.pone.0035583. 15. Muñoz M, Rodríguez MC, Alves E, Folch JM, Ibañez-escriche N, Silió L, et al. 35. Liaubet L, Lobjois V, Faraut T, Tircazes A, Benne F, Iannuccelli N, et al. Genome-wide analysis of porcine backfat and intramuscular fat fatty acid Genetic variability of transcript abundance in pig peri-mortem skeletal composition using high-density genotyping and expression data. BMC muscle: eQTL localized genes involved in stress response, cell death, muscle Genomics. 2013;14:845. disorders and metabolism. BMC Genomics. 2011;12:548. 16. Heidt H, Ulas M, Muhammad C, Uddin J, Looft C, Große-brinkhaus C. A 36. Gaffney DJ, Veyrieras J-B, Degner JF, Pique-Regi R, Pai AA, Crawford GE, et genetical genomics approach reveals new candidates and confirms known al. Dissecting the regulatory architecture of gene expression QTLs. Genome candidate genes for drip loss in a porcine resource population. Mamm Biol. 2012;13:R7. https://doi.org/10.1186/gb-2012-13-1-r7. Genome. 2013;24:416–26. 37. Bryois J, Buil A, Evans DM, Kemp JP, Montgomery SB, Conrad DF, et al. Cis 17. Steibel JP, Bates RO, Rosa GJM, Tempelman RJ, Valencia D, Ragavendran A, and trans effects of human genomic variants on gene expression. PLoS et al. Genome-wide linkage analysis of global gene expression in loin Genet. 2014;10:e1004461. https://doi.org/10.1371/journal.pgen.1004461. muscle tissue identifies candidate genes in pigs. PLoS One. 2011;6:e16766. 38. Wright FA, Sullivan PF, Brooks AI, Zou F, Sun W, Xia K, et al. Heritability and 18. Edwards DB, Ernst CW, Tempelman RJ, Rosa GJM, Raney NE, Hoge MD, et al. genomics of gene expression in peripheral blood. Nat Genet. 2014;46:430–7. Quantitative trait loci mapping in an F2 Duroc x Pietrain resource https://doi.org/10.1038/ng.2951. population: I. Growth traits. J Anim Sci. 2008;86:241–53. 39. Peñagaricano F, Valente BD, Steibel JP, Bates RO, Ernst CW, Khatib H, et al. 19. Edwards DB, Ernst CW, Raney NE, Doumit ME, Hoge MD, Bates RO. Exploring causal networks underlying fat deposition and muscularity in pigs Quantitative trait locus mapping in an F2 Duroc x Pietrain resource through the integration of phenotypic, genotypic and transcriptomic data. population: II. Carcass and meat quality traits. J Anim Sci. 2008;86:254–66. BMC Syst Biol. 2015;9:58. https://doi.org/10.1186/s12918-015-0207-6. https://doi.org/10.2527/jas.2006-626. 40. Ng MCY, Graff M, Lu Y, Justice AE, Mudgal P, Liu CT, et al. Discovery and 20. Gualdrón Duarte JL, Cantet RJC, Bernal Rubio YL, Bates RO, Ernst CW, Raney fine-mapping of adiposity loci using high density imputation of genome- NE, et al. Refining genomewide association for growth and fat deposition wide association studies in individuals of African ancestry: african ancestry traits in an F 2 pig population. J Anim Sci. 2016;94:1387–97. anthropometry genetics consortium. PLoS Genet. 2017;13:e1006719. 21. Casiró S, Ernst CW, Raney NE, Bates RO, Charles MG, Steibel JP. 41. Zhang C, Bruce H, Yang T, Charagu P, Kemp RA, Boddicker N, et al. Genome Genome-wide association study in an F2 Duroc x Pietrain resource wide association studies (GWAS) identify QTL on SSC2 and SSC17 affecting population for economically important meat quality and carcass traits. J loin peak shear force in crossbred commercial pigs. PLoS One. 2016;11: Anim Sci. 2017;95:554–8. e0145082. https://doi.org/10.1371/journal.pone.0145082. 22. Milan D, Jeon J, Looft C, Amarger V, Robic A, Thelander M, et al. A mutation in 42. Nonneman DJ, Shackelford SD, King DA, Wheeler TL, Wiedmann RT, Snelling PRKAG3 associated with excess glycogen content in pig skeletal muscle. WM, et al. Genome-wide association of meat quality traits and tenderness Science. 2000;288(80):1248–51. https://doi.org/10.1126/science.288.5469.1248. in swine. J Anim Sci. 2013;91:4043–50. 23. Ciobanu D, Bastiaansen J, Malek M, Helm J, Woollard J, Plastow G, et al. 43. Tepass U. FERM proteins in animal morphogenesis. Curr Opin Genet Dev. Evidence for new alleles in the protein kinase adenosine 2009;19:357–67. monophosphateactivated gamma3-subunit gene associated with low 44. Chang CL, Hsieh TS, Yang TT, Rothberg KG, Azizoglu DB, Volk E, et al. glycogen content in pig skeletal muscle and improved meat quality. Feedback regulation of receptor-induced ca2+ signaling mediated by e-syt1 Genetics. 2001;159:1151–62. and nir2 at endoplasmic reticulum-plasma membrane junctions. Cell Rep. 24. Guérineau NC, Desarménien MG, Carabelli V, Carbone E. Functional 2013;5:813–25. https://doi.org/10.1016/j.celrep.2013.09.038. chromaffin cell plasticity in response to stress: focus on nicotinic, gap 45. Chang CL, Liou J. Phosphatidylinositol 4, 5-bisphosphate homeostasis junction, and voltage-gated Ca2+ channels. J Mol Neurosci. 2012;48:368–86. regulated by Nir2 and Nir3 proteins at endoplasmic reticulum-plasma 25. Di Cesare ML, Cinci L, Micheli L, Zanardelli M, Pacini A, McIntosh JM, et al. α- membrane junctions. J Biol Chem. 2015;290:14289–301. Conotoxin RgIA protects against the development of nerve injury-induced 46. Chang CL, Liou J. Homeostatic regulation of the PI(4,5)P2-Ca2+ signaling chronic pain and prevents both neuronal and glial derangement. Pain. system at ER-PM junctions. Biochim Biophys Acta - Mol Cell Biol Lipids. 2014;155:1986–95. 1861;2016:862–73. Velez-Irizarry et al. BMC Genomics (2019) 20:3 Page 19 of 19

47. Kim NK, Cho S, Lee SH, Park HR, Lee CS, Cho YM, et al. Proteins in 69. Cardoso FF, Rosa GJM, Steibel JP, Ernst CW, Bates RO, Tempelman RJ. longissimus muscle of Korean native cattle and their relationship to meat Selective transcriptional profiling and data analysis strategies for expression quality. Meat Sci. 2008;80:1068–73. quantitative trait loci mapping in outbred F2 populations. Genetics. 2008; 48. Berridge MJ. Inositol trisphosphate and calcium signalling mechanisms. 180:1679–90. https://doi.org/10.1534/genetics.108.090969. Biochim Biophys Acta - Mol Cell Res. 1793;2009:933–40. https://doi.org/10. 70. Gualdrón Duarte JL, Bates RO, Ernst CW, Raney NE, Cantet RJC, Steibel JP. 1016/j.bbamcr.2008.10.005. Genotype imputation accuracy in a F2 pig population using high density 49. Csernoch L, Jacquemond V. Phosphoinositides in Ca2+ signaling and and low density SNP panels. BMC Genet. 2013;14:38 http://www. excitation–contraction coupling in skeletal muscle: an old player and pubmedcentral.nih.gov/articlerender.fcgi?artid=3655050&tool= newcomers. J Muscle Res Cell Motil. 2015;36:491–9. pmcentrez&rendertype=abstract. 50. Anderson DM, Makarewich CA, Anderson KM, Shelton JM, Bezprozvannaya 71. Badke YM, Bates RO, Ernst CW, Schwab C, Fix J, Van Tassell CP, et al. S, Bassel-Duby R, et al. Widespread control of calcium signaling by a family Methods of tagSNP selection and other variables affecting imputation of SERCA-inhibiting micropeptides. Sci Signal. 2016;9:ra119. accuracy in swine. BMC Genet. 2013;14:8 http://www.pubmedcentral.nih. 51. Cerrone M, Montnach J, Lin X, Zhao YT, Zhang M, Agullo-Pascual E, et al. gov/articlerender.fcgi?artid=3734000&tool=pmcentrez&rendertype=abstract. Plakophilin-2 is required for transcription of genes that control calcium 72. Ramos AM, Crooijmans RPMA, Affara NA, Amaral AJ, Archibald AL, Beever JE, cycling and cardiac rhythm. Nat Commun. 2017;8:106. et al. Design of a high density SNP genotyping assay in the pig using SNPs 52. Lu L, Chen Y, Zhu Y. The molecular basis of targeting PFKFB3 as a identified and characterized by next generation sequencing technology. therapeutic strategy against cancer. Oncotarget. 2017;8:62793–802. PLoS One. 2009;4:e6524. https://doi.org/10.1371/journal.pone.0006524. 53. Yalcin A, Clem BF, Imbert-Fernandez Y, Ozcan SC, Peker S, O’Neal J, et al. 73. Bolger AM, Lohse M, Usadel B. Trimmomatic: a flexible trimmer for Illumina 6-Phosphofructo-2-kinase (PFKFB3) promotes cell cycle progression and sequence data. Bioinformatics. 2014;30:2114–20. suppresses apoptosis via Cdk1-mediated phosphorylation of p27. Cell Death Dis. 74. Gordon A, Hannon GJ. Fastx-toolkit. Computer program distributed by the 2014;5:e1337. author 2010. http://hannonlab.cshl.edu/fastx_toolkit/index.html. 54. Li F, Jin D, Tang C, Gao D. CEP55 promotes cell proliferation and inhibits 75. Kim D, Pertea G, Trapnell C, Pimentel H, Kelley R, Salzberg SL. TopHat2: apoptosis via the pi3k/akt/p21 signaling pathway in human glioma u251 accurate alignment of transcriptomes in the presence of insertions, cells. Oncol Lett. 2018;15:4789–96. deletions and gene fusions. Genome Biol. 2013;14:R36. 55. Park HY, Lee S-B, Yoo H-Y, Kim S-J, Kim W-S, Kim J-I, et al. Whole-exome and 76. Trapnell C, Pachter L, Salzberg SL. TopHat: discovering splice junctions with – transcriptome sequencing of refractory diffuse large B-cell lymphoma. RNA-Seq. Bioinformatics. 2009;25:1105 11. https://doi.org/10.1093/ Oncotarget. 2016;7:86433–45. https://doi.org/10.18632/oncotarget.13239. bioinformatics/btp120. 56. Park-Windhol C, Ng YS, Yang J, Primo V, Saint-Geniez M, D’amore PA. 77. Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, et al. The Endomucin inhibits VEGF-induced endothelial cell migration, growth, and sequence alignment/map format and SAMtools. Bioinformatics. 2009;25: – morphogenesis by modulating VEGFR2 signaling. Sci Rep. 2017;7:17138. 2078 9. https://doi.org/10.1093/bioinformatics/btp352. 57. Lin C, McGough R, Aswad B, Block J, Terek R. Hypoxia induces HIF-1-alpha 78. Anders S, Pyl PT, Huber W. HTSeq a Python framework to work with high- – and VEGF expression in chondrosarcoma cells and chondrocytes. throughput sequencing data. Bioinformatics. 2014;31:166 9. https://doi.org/ J Orthop Res. 2004;22:1175–81. 10.1101/002824. 58. Lampiasi N, Montana G. An in vitro inflammation model to study the Nrf2 79. Robinson MD, Oshlack A. A scaling normalization method for differential and NF-κB crosstalk in presence of ferulic acid as modulator. Immunobiology. expression analysis of RNA-seq data. Genome Biol. 2010;11:R25. 2017;223:339–55. https://doi.org/10.1016/j.imbio.2017.10.046. 80. Dillies M-A, Rau A, Aubert J, Hennequet-Antier C, Jeanmougin M, Servant N, 59. Chen H, Tang X, Zhou B, Zhou Z, Xu N, Wang Y. A ROS-mediated et al. A comprehensive evaluation of normalization methods for Illumina mitochondrial pathway and Nrf2 pathway activation are involved in BDE-47 high-throughput RNA sequencing data analysis. Brief Bioinform. 2013;14: – induced apoptosis in neuro-2a cells. Chemosphere. 2017;184:679–86. 671 83. https://doi.org/10.1093/bib/bbs046. https://doi.org/10.1016/j.chemosphere.2017.06.006. 81. Law CW, Chen Y, Shi W, Smyth GK. Voom: precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol. 2014;15:R29. 60. Miller CJ, Gounder SS, Kannan S, Goutam K, Muthusamy VR, Firpo MA, et al. https://doi.org/10.1186/gb-2014-15-2-r29. Disruption of Nrf2/ARE signaling impairs antioxidant mechanisms and 82. Vanraden PM. Efficient methods to compute genomic predictions. promotes cell degradation pathways in aged skeletal muscle. Biochim J Dairy Sci. 2008;91:4414–23. https://doi.org/10.3168/jds.2007-0980. Biophys Acta. 1822;2012:1038–50. 83. Visscher PM. A note on the asymptotic distribution of likelihood ratio tests 61. Horie M, Warabi E, Komine S, Oh S, Shoda J. Cytoprotective role of Nrf2 in to test variance components. Twin Res Hum Genet. 2006;9:490–5. electrical pulse stimulated C2C12 myotube. PLoS One. 2015;10:e0144835. 84. Benjamini Y, Hochberg Y. Controlling the false discovery rate : a practical 62. El-Hattab AW, Scaglia F. Gene reviews: SUCLG1-related mitochondrial DNA and powerful approach to multiple testing. J R Stat Soc. 1995;57:289–300. depletion syndrome, Encephalomyopathic form with Methylmalonic 85. Bauer DF. Constructing confidence sets using rank statistics. J Am Stat Assoc. aciduria. Seattle: University of Washington, Seattle; 2017. https://www.ncbi. 1972;67:687–90. nlm.nih.gov/books/NBK425223/ 86. Gualdrón Duarte JL, Cantet RJC, Bates RO, Ernst CW, Raney NE, Steibel JP. 63. Miller C, Wang L, Ostergaard E, Dan P, Saada A. The interplay between Rapid screening for phenotype-genotype associations by linear SUCLA2, SUCLG2, and mitochondrial DNA depletion. Biochim Biophys Acta. transformations of genomic evaluations. BMC Bioinformatics. 2014;15:246. 1812;2011:625–9. https://doi.org/10.1016/j.bbadis.2011.01.013. 87. Bernal Rubio YL, Guardrón Duarte JL, Bates RO, Ernst CW, Nonneman D, 64. Horton JS, Wakano CT, Speck M, Stokes AJ. Two-pore channel 1 Rohrer GA, et al. Implementing meta-analysis from genome-wide interacts with citron kinase, regulating completion of cytokinesis. association studies for pork quality traits 1. J Anim Sci. 2015;93:5607–17. Channels. 2015;9:21–9. 88. KangHM,ZaitlenNA,WadeCM,KirbyA,HeckermanD,DalyMJ,etal. 65. Gregorich ZR, Peng Y, Cai W, Jin Y, Wei L, Chen AJ, et al. Top-down Efficient control of population structure in model organism association targeted proteomics reveals decrease in myosin regulatory light-chain mapping. Genetics. 2008;178:1709– 23. https://doi.org/10.1534/genetics. phosphorylation that contributes to sarcopenic muscle dysfunction. 107.080101. – J Proteome Res. 2016;15:2706 16. 89. Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Paepe A, 66. Rose A, Schraegle SJ, Stahlberg EA, Meier I. Coiled-coil protein composition et al. Accurate normalization of real-time quantitative RT-PCR data by of 22 proteomes - differences and common themes in subcellular geometric averaging of multiple internal control genes. Genome Biol. infrastructure and traffic control. BMC Evol Biol. 2005;5:66. 2002;3:34–1. https://doi.org/10.1186/gb-2002-3-7-research0034. 67. Kirov G, Zaharieva I, Georgieva L, Moskvina V, Nikolov I, Cichon S, et al. A genome-wide association study in 574 schizophrenia trios using DNA pooling. Mol Psychiatry. 2009;14:796–803. 68. Martins-De-Souza D, Gattaz WF, Schmitt A, Rewerts C, Maccarrone G, Dias-Neto E, et al. Prefrontal cortex shotgun proteome analysis reveals altered calcium homeostasis and immune system imbalance in schizophrenia. Eur Arch Psychiatry Clin Neurosci. 2009;259:151–63.